Author: Uri Blass
Date: 21:58:49 02/21/01
Go up one level in this thread
On February 22, 2001 at 00:41:55, Uri Blass wrote: >On February 21, 2001 at 23:37:46, jonas cohonas wrote: > >>>>> >>>>>With Deep Shredder and Deep SOS out of its way Nimzo 8 only have a real tough >>>>>Opponent left Deep Fritz, But since Nimzo 8 behave differently when using a >>>>>Processor higher than 800 MHz and it was proven in the Cadaque's Tourney and in >>>>>my private testing of several games using my AMD Athlon 800 MHz >>>> >>>> >>>> >>>>Nothing like that has been ever proved, except in your imagination. >>>> >>>> >>>> >>>> Christophe >>>> >>>> >>>Yes, why would Nimzo "behave differently" just because you took it from a 600 >>>(as an example) to an 800 Mhz cpu? Preposterous is a relevant word here, I >>>believe. >>> >>>Chuck >>> >>It is a well known fact that a program like Gandalf_432* plays differently on a >>high powered cpu than it does on a 500mhz >>So why should it be impossible for nimzo 8?? >>In my testing period of Gandalf_432f-h i have gone from a 500mhz amd k6-2 >>to a 900 mhz T-bird to a 1.2ghz T-bird, and now i just clocked my 1.2 to 1.3ghz >>in that period of time i have watched Gandalf change style (to the better) each >>time i did a cpu upgrade. For example it's almost as good at blitz now as Fritz >>6 at FICS it has a 50% score against a Deep Fritz account running on a dual 1ghz >>Pentium3! >>My point is that if this is possible for Gandalf then why not for Nimzo8??? >> >>Regards >>Jonas > >Gandalf earns more nodes from fast hardware. > >My experience is that Gandalf is more than twice faster on PIII800 than pIII450 >when other programs like Junior and Fritz are not. > >The question is if the situation with Nimzo is similiar. >There may be also another factor and maybe with the same number of nodes it >searches in a different way in fast hardware so I think that the best test is >simply to compare the result in test suites. > >I suggest the follwing test > >Give Nimzo8 a test suite that it can solve all of the positions but need some >time for it and calculate the sum of the times of the solutions. > >Do the test on different computers and compare the results. > >Do the same for other programs and you can find which program earns more from >fast hardware. > >Uri There is also a problem with this idea because some programs like Junior analyze in a different way in different time control and the sum of the times is meaningless. I think that the only way is to find eqvivalent times. If you need to give Nimzo 58-62 seconds per position on slow hardware in order to get 80 solutuions out of 100 and you need to give nimzo 29-31 seconds on fast hardware to get the same result then it means that it is clear that the fast hardware is twice faster but when I think about it then there may be also a problem. It is possible that a program does some positional evaluation only on fast hardware because the same evaluation on slow hardware costs a lot of cycles clocks when the evaluation on fast hardware is cheaper. test suites are usually tactical tests so they cannot discover it. I think that the best thing is to ask the programmers if they do something different with fast hardware(they may do it if they try to optimize their pogram for all hardwares). Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.